3 * Bacula Director Job processing routines
5 * Kern Sibbald, October MM
10 Copyright (C) 2000-2003 Kern Sibbald and John Walker
12 This program is free software; you can redistribute it and/or
13 modify it under the terms of the GNU General Public License as
14 published by the Free Software Foundation; either version 2 of
15 the License, or (at your option) any later version.
17 This program is distributed in the hope that it will be useful,
18 but WITHOUT ANY WARRANTY; without even the implied warranty of
19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 General Public License for more details.
22 You should have received a copy of the GNU General Public
23 License along with this program; if not, write to the Free
24 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
32 /* Forward referenced subroutines */
33 static void *job_thread(void *arg);
34 static char *edit_run_codes(JCR *jcr, char *omsg, char *imsg);
35 static int acquire_resource_locks(JCR *jcr);
37 static void backoff_resource_locks(JCR *jcr, int count);
38 static void release_resource_locks(JCR *jcr);
41 /* Exported subroutines */
44 /* Imported subroutines */
45 extern void term_scheduler();
46 extern void term_ua_server();
47 extern int do_backup(JCR *jcr);
48 extern int do_admin(JCR *jcr);
49 extern int do_restore(JCR *jcr);
50 extern int do_verify(JCR *jcr);
53 static semlock_t job_lock;
54 static pthread_mutex_t mutex;
55 static pthread_cond_t resource_wait;
56 static int waiting = 0; /* count of waiting threads */
63 void init_job_server(int max_workers)
67 if ((stat = sem_init(&job_lock, max_workers)) != 0) {
68 Emsg1(M_ABORT, 0, _("Could not init job lock: ERR=%s\n"), strerror(stat));
70 if ((stat = pthread_mutex_init(&mutex, NULL)) != 0) {
71 Emsg1(M_ABORT, 0, _("Could not init resource mutex: ERR=%s\n"), strerror(stat));
73 if ((stat = pthread_cond_init(&resource_wait, NULL)) != 0) {
74 Emsg1(M_ABORT, 0, _("Could not init resource wait: ERR=%s\n"), strerror(stat));
79 if ((stat = jobq_init(&job_queue, max_workers, job_thread)) != 0) {
80 Emsg1(M_ABORT, 0, _("Could not init job queue: ERR=%s\n"), strerror(stat));
88 * Run a job -- typically called by the scheduler, but may also
89 * be called by the UA (Console program).
92 void run_job(JCR *jcr)
99 sm_check(__FILE__, __LINE__, True);
100 init_msg(jcr, jcr->messages);
101 create_unique_job_name(jcr, jcr->job->hdr.name);
102 set_jcr_job_status(jcr, JS_Created);
103 jcr->jr.SchedTime = jcr->sched_time;
104 jcr->jr.StartTime = jcr->start_time;
105 jcr->jr.Type = jcr->JobType;
106 jcr->jr.Level = jcr->JobLevel;
107 jcr->jr.JobStatus = jcr->JobStatus;
108 bstrncpy(jcr->jr.Name, jcr->job->hdr.name, sizeof(jcr->jr.Name));
109 bstrncpy(jcr->jr.Job, jcr->Job, sizeof(jcr->jr.Job));
111 /* Initialize termination condition variable */
112 if ((errstat = pthread_cond_init(&jcr->term_wait, NULL)) != 0) {
113 Jmsg1(jcr, M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
114 set_jcr_job_status(jcr, JS_ErrorTerminated);
122 Dmsg0(50, "Open database\n");
123 jcr->db=db_init_database(jcr, jcr->catalog->db_name, jcr->catalog->db_user,
124 jcr->catalog->db_password, jcr->catalog->db_address,
125 jcr->catalog->db_port, jcr->catalog->db_socket);
126 if (!db_open_database(jcr, jcr->db)) {
127 Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
128 set_jcr_job_status(jcr, JS_ErrorTerminated);
132 Dmsg0(50, "DB opened\n");
137 jcr->jr.JobStatus = jcr->JobStatus;
138 if (!db_create_job_record(jcr, jcr->db, &jcr->jr)) {
139 Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
140 set_jcr_job_status(jcr, JS_ErrorTerminated);
144 jcr->JobId = jcr->jr.JobId;
145 ASSERT(jcr->jr.JobId > 0);
147 Dmsg4(30, "Created job record JobId=%d Name=%s Type=%c Level=%c\n",
148 jcr->JobId, jcr->Job, jcr->jr.Type, jcr->jr.Level);
149 Dmsg0(200, "Add jrc to work queue\n");
152 if ((stat = pthread_create(&tid, NULL, job_thread, (void *)jcr)) != 0) {
153 Emsg1(M_ABORT, 0, _("Unable to create job thread: ERR=%s\n"), strerror(stat));
157 /* Queue the job to be run */
158 if ((stat = jobq_add(&job_queue, jcr)) != 0) {
159 Emsg1(M_ABORT, 0, _("Could not add job queue: ERR=%s\n"), strerror(stat));
163 Dmsg0(200, "Done run_job()\n");
167 * This is the engine called by workq_add() when we were pulled
168 * from the work queue.
169 * At this point, we are running in our own thread
171 static void *job_thread(void *arg)
173 JCR *jcr = (JCR *)arg;
175 pthread_detach(pthread_self());
176 sm_check(__FILE__, __LINE__, True);
179 if (!acquire_resource_locks(jcr)) {
180 set_jcr_job_status(jcr, JS_Canceled);
183 Dmsg0(200, "=====Start Job=========\n");
184 jcr->start_time = time(NULL); /* set the real start time */
185 set_jcr_job_status(jcr, JS_Running);
187 if (job_canceled(jcr)) {
188 update_job_end_record(jcr);
189 } else if (jcr->job->MaxStartDelay != 0 && jcr->job->MaxStartDelay <
190 (utime_t)(jcr->start_time - jcr->sched_time)) {
191 Jmsg(jcr, M_FATAL, 0, _("Job canceled because max start delay time exceeded.\n"));
192 set_jcr_job_status(jcr, JS_Canceled);
193 update_job_end_record(jcr);
197 if (jcr->job->RunBeforeJob) {
198 POOLMEM *before = get_pool_memory(PM_FNAME);
201 char line[MAXSTRING];
203 before = edit_run_codes(jcr, before, jcr->job->RunBeforeJob);
204 bpipe = open_bpipe(before, 0, "r");
205 free_pool_memory(before);
206 while (fgets(line, sizeof(line), bpipe->rfd)) {
207 Jmsg(jcr, M_INFO, 0, _("RunBefore: %s"), line);
209 status = close_bpipe(bpipe);
211 Jmsg(jcr, M_FATAL, 0, _("RunBeforeJob returned non-zero status=%d\n"),
213 set_jcr_job_status(jcr, JS_FatalError);
214 update_job_end_record(jcr);
218 switch (jcr->JobType) {
221 if (jcr->JobStatus == JS_Terminated) {
227 if (jcr->JobStatus == JS_Terminated) {
233 if (jcr->JobStatus == JS_Terminated) {
239 if (jcr->JobStatus == JS_Terminated) {
244 Pmsg1(0, "Unimplemented job type: %d\n", jcr->JobType);
247 if (jcr->job->RunAfterJob) {
248 POOLMEM *after = get_pool_memory(PM_FNAME);
251 char line[MAXSTRING];
253 after = edit_run_codes(jcr, after, jcr->job->RunAfterJob);
254 bpipe = open_bpipe(after, 0, "r");
255 free_pool_memory(after);
256 while (fgets(line, sizeof(line), bpipe->rfd)) {
257 Jmsg(jcr, M_INFO, 0, _("RunAfter: %s"), line);
259 status = close_bpipe(bpipe);
261 Jmsg(jcr, M_FATAL, 0, _("RunAfterJob returned non-zero status=%d\n"),
263 set_jcr_job_status(jcr, JS_FatalError);
264 update_job_end_record(jcr);
270 release_resource_locks(jcr);
271 if (jcr->job->RescheduleOnError &&
272 jcr->JobStatus != JS_Terminated &&
273 jcr->JobStatus != JS_Canceled &&
274 jcr->job->RescheduleTimes > 0 &&
275 jcr->reschedule_count < jcr->job->RescheduleTimes) {
278 * Reschedule this job by cleaning it up, but
279 * reuse the same JobId if possible.
281 jcr->reschedule_count++;
282 jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
283 Dmsg2(100, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job,
284 (int)jcr->job->RescheduleInterval);
285 jcr->JobStatus = JS_Created; /* force new status */
286 dird_free_jcr(jcr); /* partial cleanup old stuff */
287 if (jcr->JobBytes == 0) {
288 continue; /* reschedule the job */
291 * Something was actually backed up, so we cannot reuse
292 * the old JobId or there will be database record
293 * conflicts. We now create a new job, copying the
294 * appropriate fields.
296 JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
297 set_jcr_defaults(njcr, jcr->job);
298 njcr->reschedule_count = jcr->reschedule_count;
299 njcr->JobLevel = jcr->JobLevel;
300 njcr->JobStatus = jcr->JobStatus;
301 njcr->pool = jcr->pool;
302 njcr->store = jcr->store;
303 njcr->messages = jcr->messages;
312 Dmsg0(200, "Close DB\n");
313 db_close_database(jcr, jcr->db);
318 Dmsg0(50, "======== End Job ==========\n");
319 sm_check(__FILE__, __LINE__, True);
324 * Acquire the resources needed. These locks limit the
325 * number of jobs by each resource. We have limits on
326 * Jobs, Clients, Storage, and total jobs.
328 static int acquire_resource_locks(JCR *jcr)
331 time_t now = time(NULL);
332 time_t wtime = jcr->sched_time - now;
334 /* Wait until scheduled time arrives */
335 if (wtime > 0 && verbose) {
336 Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"),
338 set_jcr_job_status(jcr, JS_WaitStartTime);
340 /* Check every 30 seconds if canceled */
342 Dmsg2(100, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime);
346 bmicrosleep(wtime, 0);
347 if (job_canceled(jcr)) {
350 wtime = jcr->sched_time - time(NULL);
358 /* Initialize semaphores */
359 if (jcr->store->sem.valid != SEMLOCK_VALID) {
360 if ((stat = sem_init(&jcr->store->sem, jcr->store->MaxConcurrentJobs)) != 0) {
361 Emsg1(M_ABORT, 0, _("Could not init Storage semaphore: ERR=%s\n"), strerror(stat));
364 if (jcr->client->sem.valid != SEMLOCK_VALID) {
365 if ((stat = sem_init(&jcr->client->sem, jcr->client->MaxConcurrentJobs)) != 0) {
366 Emsg1(M_ABORT, 0, _("Could not init Client semaphore: ERR=%s\n"), strerror(stat));
369 if (jcr->job->sem.valid != SEMLOCK_VALID) {
370 if ((stat = sem_init(&jcr->job->sem, jcr->job->MaxConcurrentJobs)) != 0) {
371 Emsg1(M_ABORT, 0, _("Could not init Job semaphore: ERR=%s\n"), strerror(stat));
376 /* Acquire semaphore */
377 set_jcr_job_status(jcr, JS_WaitJobRes);
378 if ((stat = sem_lock(&jcr->job->sem)) != 0) {
379 Emsg1(M_ABORT, 0, _("Could not acquire Job max jobs lock: ERR=%s\n"), strerror(stat));
381 set_jcr_job_status(jcr, JS_WaitClientRes);
382 if ((stat = sem_trylock(&jcr->client->sem)) != 0) {
384 backoff_resource_locks(jcr, 1);
387 Emsg1(M_ABORT, 0, _("Could not acquire Client max jobs lock: ERR=%s\n"), strerror(stat));
390 set_jcr_job_status(jcr, JS_WaitStoreRes);
391 if ((stat = sem_trylock(&jcr->store->sem)) != 0) {
393 backoff_resource_locks(jcr, 2);
396 Emsg1(M_ABORT, 0, _("Could not acquire Storage max jobs lock: ERR=%s\n"), strerror(stat));
399 set_jcr_job_status(jcr, JS_WaitMaxJobs);
400 if ((stat = sem_trylock(&job_lock)) != 0) {
402 backoff_resource_locks(jcr, 3);
405 Emsg1(M_ABORT, 0, _("Could not acquire max jobs lock: ERR=%s\n"), strerror(stat));
411 if (job_canceled(jcr)) {
416 * Wait for a resource to be released either by backoff or
417 * by a job terminating.
420 pthread_cond_wait(&resource_wait, &mutex);
425 jcr->acquired_resource_locks = true;
432 * We could not get all the resource locks because
433 * too many jobs are running, so release any locks
434 * we did acquire, giving others a chance to use them
437 static void backoff_resource_locks(JCR *jcr, int count)
442 sem_unlock(&jcr->store->sem);
443 /* Fall through wanted */
445 sem_unlock(&jcr->client->sem);
446 /* Fall through wanted */
448 sem_unlock(&jcr->job->sem);
452 * Since we released a lock, if there are any threads
453 * waiting, wake them up so that they can try again.
456 pthread_cond_broadcast(&resource_wait);
463 * This is called at the end of the job to release
464 * any resource limits on the number of jobs. If
465 * there are any other jobs waiting, we wake them
466 * up so that they can try again.
469 static void release_resource_locks(JCR *jcr)
471 if (!jcr->acquired_resource_locks) {
472 return; /* Job canceled, no locks acquired */
475 sem_unlock(&jcr->store->sem);
476 sem_unlock(&jcr->client->sem);
477 sem_unlock(&jcr->job->sem);
478 sem_unlock(&job_lock);
480 pthread_cond_broadcast(&resource_wait);
482 jcr->acquired_resource_locks = false;
488 * Get or create a Client record for this Job
490 int get_or_create_client_record(JCR *jcr)
494 memset(&cr, 0, sizeof(cr));
495 bstrncpy(cr.Name, jcr->client->hdr.name, sizeof(cr.Name));
496 cr.AutoPrune = jcr->client->AutoPrune;
497 cr.FileRetention = jcr->client->FileRetention;
498 cr.JobRetention = jcr->client->JobRetention;
499 if (!jcr->client_name) {
500 jcr->client_name = get_pool_memory(PM_NAME);
502 pm_strcpy(&jcr->client_name, jcr->client->hdr.name);
503 if (!db_create_client_record(jcr, jcr->db, &cr)) {
504 Jmsg(jcr, M_FATAL, 0, _("Could not create Client record. ERR=%s\n"),
505 db_strerror(jcr->db));
508 jcr->jr.ClientId = cr.ClientId;
510 if (!jcr->client_uname) {
511 jcr->client_uname = get_pool_memory(PM_NAME);
513 pm_strcpy(&jcr->client_uname, cr.Uname);
515 Dmsg2(100, "Created Client %s record %d\n", jcr->client->hdr.name,
522 * Write status and such in DB
524 void update_job_end_record(JCR *jcr)
526 if (jcr->jr.EndTime == 0) {
527 jcr->jr.EndTime = time(NULL);
529 jcr->end_time = jcr->jr.EndTime;
530 jcr->jr.JobId = jcr->JobId;
531 jcr->jr.JobStatus = jcr->JobStatus;
532 jcr->jr.JobFiles = jcr->JobFiles;
533 jcr->jr.JobBytes = jcr->JobBytes;
534 jcr->jr.VolSessionId = jcr->VolSessionId;
535 jcr->jr.VolSessionTime = jcr->VolSessionTime;
536 if (!db_update_job_end_record(jcr, jcr->db, &jcr->jr)) {
537 Jmsg(jcr, M_WARNING, 0, _("Error updating job record. %s"),
538 db_strerror(jcr->db));
543 * Takes base_name and appends (unique) current
544 * date and time to form unique job name.
546 * Returns: unique job name in jcr->Job
547 * date/time in jcr->start_time
549 void create_unique_job_name(JCR *jcr, char *base_name)
551 /* Job start mutex */
552 static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
553 static time_t last_start_time = 0;
556 char dt[MAX_TIME_LENGTH];
557 char name[MAX_NAME_LENGTH];
560 /* Guarantee unique start time -- maximum one per second, and
561 * thus unique Job Name
563 P(mutex); /* lock creation of jobs */
565 while (now == last_start_time) {
566 bmicrosleep(0, 500000);
569 last_start_time = now;
570 V(mutex); /* allow creation of jobs */
571 jcr->start_time = now;
572 /* Form Unique JobName */
573 localtime_r(&now, &tm);
574 /* Use only characters that are permitted in Windows filenames */
575 strftime(dt, sizeof(dt), "%Y-%m-%d_%H.%M.%S", &tm);
576 bstrncpy(name, base_name, sizeof(name));
577 name[sizeof(name)-22] = 0; /* truncate if too long */
578 bsnprintf(jcr->Job, sizeof(jcr->Job), "%s.%s", name, dt); /* add date & time */
579 /* Convert spaces into underscores */
580 for (p=jcr->Job; *p; p++) {
588 * Free the Job Control Record if no one is still using it.
589 * Called from main free_jcr() routine in src/lib/jcr.c so
590 * that we can do our Director specific cleanup of the jcr.
592 void dird_free_jcr(JCR *jcr)
594 Dmsg0(200, "Start dird free_jcr\n");
596 if (jcr->sd_auth_key) {
597 free(jcr->sd_auth_key);
598 jcr->sd_auth_key = NULL;
604 if (jcr->file_bsock) {
605 Dmsg0(200, "Close File bsock\n");
606 bnet_close(jcr->file_bsock);
607 jcr->file_bsock = NULL;
609 if (jcr->store_bsock) {
610 Dmsg0(200, "Close Store bsock\n");
611 bnet_close(jcr->store_bsock);
612 jcr->store_bsock = NULL;
615 Dmsg0(200, "Free JCR fname\n");
616 free_pool_memory(jcr->fname);
620 Dmsg0(200, "Free JCR stime\n");
621 free_pool_memory(jcr->stime);
624 if (jcr->RestoreBootstrap) {
625 free(jcr->RestoreBootstrap);
626 jcr->RestoreBootstrap = NULL;
628 if (jcr->client_uname) {
629 free_pool_memory(jcr->client_uname);
630 jcr->client_uname = NULL;
632 Dmsg0(200, "End dird free_jcr\n");
636 * Set some defaults in the JCR necessary to
637 * run. These items are pulled from the job
638 * definition as defaults, but can be overridden
639 * later either by the Run record in the Schedule resource,
640 * or by the Console program.
642 void set_jcr_defaults(JCR *jcr, JOB *job)
645 jcr->JobType = job->JobType;
646 jcr->JobLevel = job->level;
647 jcr->JobPriority = job->Priority;
648 jcr->store = job->storage;
649 jcr->client = job->client;
650 if (!jcr->client_name) {
651 jcr->client_name = get_pool_memory(PM_NAME);
653 pm_strcpy(&jcr->client_name, jcr->client->hdr.name);
654 jcr->pool = job->pool;
655 jcr->catalog = job->client->catalog;
656 jcr->fileset = job->fileset;
657 jcr->messages = job->messages;
658 if (jcr->RestoreBootstrap) {
659 free(jcr->RestoreBootstrap);
661 /* This can be overridden by Console program */
662 if (job->RestoreBootstrap) {
663 jcr->RestoreBootstrap = bstrdup(job->RestoreBootstrap);
665 /* If no default level given, set one */
666 if (jcr->JobLevel == 0) {
667 switch (jcr->JobType) {
669 jcr->JobLevel = L_VERIFY_CATALOG;
672 jcr->JobLevel = L_INCREMENTAL;
676 jcr->JobLevel = L_FULL;
685 * Edit codes into Run command
688 * %d = Director's name
696 * omsg = edited output message
697 * imsg = input string containing edit codes (%x)
700 static char *edit_run_codes(JCR *jcr, char *omsg, char *imsg)
707 Dmsg1(200, "edit_run_codes: %s\n", imsg);
708 for (p=imsg; *p; p++) {
715 str = jcr->client_name;
724 str = job_status_to_str(jcr->JobStatus);
727 sprintf(add, "%d", jcr->JobId);
734 str = job_level_to_str(jcr->JobLevel);
737 str = jcr->job->hdr.name;
740 str = job_type_to_str(jcr->JobType);
754 Dmsg1(200, "add_str %s\n", str);
755 pm_strcat(&omsg, (char *)str);
756 Dmsg1(200, "omsg=%s\n", omsg);